Wavefront-coded image restoration technology based on multi-scale deep neural networks

Optical imaging systems are significantly affected by aerodynamic thermal effects under varying flight conditions, resulting in complex image blurring. To address this challenge, this study proposes a novel wavefront-coded image restoration method based on a multi-scale deep autoencoder neural network (MS-DAE). By modulating blur levels and incorporating a multi-scale loss function with residual attention mechanisms, the proposed method achieves a remarkable improvement in peak signal-to-noise ratio (PSNR) by 16.03 dB and structural similarity index measure (SSIM) by 0.3834 compared to Wiener filtering; compared to the BaseNet model referenced in this paper, the proposed model achieves improvements of 5.69 dB and 0.026 in PSNR and SSIM, respectively, and demonstrates superior performance in detail restoration in the reconstructed images. The approach effectively restores image details, suppresses artifacts, and adapts to diverse flight conditions, demonstrating significant potential for practical applications.